How to accelerate the convergence for optimization of MECI

[panwangzhou]

Dear all,

Recently I used the XMS-CASPT2 to optimize the MECI between S0 and S1 of a molecule with BAGEL, I have sucessfully located the MECI at SA2-CASSCF level and used it as the start geometry for the XMS-CASPT2 calculation. After 77 cycles I found that there is no good convergence trend, then I kill the job, change the maxstep to 0.01, reduce the convergen threshold, and resubmit the job. However, the convergence trend did not change significantly. So I want to know which parameters are helpful for MECI optimization? The optimization method, the method of updating hessian, or the method for calcualting NACME, etc.

Part of input file:
{
"title" : "optimize",
"opttype" : "meci",
"target" : 0,
"target2" : 1,
"maxziter" : 200,
"maxstep" : 0.01,
"maxgrad" : 0.003,
"maxdisp" : 0.012,
"maxchange" : 0.00003,
"molden" : "true",
"method" : [ {
"title" : "caspt2",
"smith" : {
"method" : "caspt2",
"ms" : "true",
"xms" : "true",
"sssr" : "true",
"shift" : 0.2,
"frozen" : true,
"maxiter" : 100
},
"charge" : 0,
"nstate" : 2,
"nact" : 8,
"nclosed" : 86,
"natocc" : "true",
"maxiter" : 200,
"restart_cas" : "true"
} ]
},

The last of the outpuf file:
*** Conical intersection optimization started ***
iter energy gap energy grad rms time

  1      -1373.17543386          0.00003656          0.00351362     2903.75
  2      -1373.17520381          0.00120217          0.00361351     3029.68
  3      -1373.17429411          0.00101771          0.00342756     2951.25
  4      -1373.17478220          0.00049427          0.00300190     3202.36
  5      -1373.17560701          0.00061631          0.00293842     3156.61
  6      -1373.17539582          0.00022999          0.00534487     3344.72